Sie sind auf Seite 1von 32

J. Fluid Mech. (2004), vol. 504, pp. 239270.

c
2004 Cambridge University Press
DOI: 10.1017/S0022112004008274 Printed in the United Kingdom
239
A study of shock waves in expanding ows on
the basis of spectroscopic experiments and
quasi-gasdynamic equations
By I. A. GRAUR
1
, T. G. ELI ZAROVA
1
, A. RAMOS
2
,
G. TEJ EDA
2
, J. M. FERN

ANDEZ
2
AND S. MONTERO
2

1
Institute for Mathematical Modeling, RAS Miusskaya Sq., 4a, 125047 Moscow, Russia
2
Instituto de Estructura de la Materia, CSIC Serrano 121, 28006 Madrid, Spain
(Received 24 April 2003 and in revised form 20 October 2003)
A comprehensive numerical and experimental study of normal shock waves in
hypersonic axisymmetric jets of N
2
is presented. The numerical interpretation is
based on the quasi-gasdynamic (QGD) approach, and its generalization (QGDR) for
the breakdown of rotationaltranslational equilibrium. The experimental part, based
on diagnostics by high-sensitivity Raman spectroscopy, provides absolute density and
rotational temperatures along the expansion axis, including the wake beyond the
shock. These quantities are used as a reference for the numerical work. The limits
of applicability of the QGD approach in terms of the local Knudsen number, the
inuence of the computational grid on the numerical solution, the breakdown of
rotationtranslation equilibrium, and the possible formation of a recirculation vortex
immediately downstream from the normal shock wave are the main topics considered.
1. Introduction
Shock waves associated with the two-dimensional axisymmetric expansion of a
gas show, more or less modied, many of the well-known peculiarities of the one-
dimensional expansion, for instance the sharp density and temperature gradients
across the normal shock wave, the breakdown of the thermodynamic equilibrium
between translation and rotation degrees of freedom, the lag between the proles of
density, translational temperature and rotational temperature, the thermal overshoot
at the end of the shock wave, and the bimodal distribution of temperatures across it.
In addition, a characteristic of two-dimensional axisymmetric supersonic expansions
seems to be a recirculation vortex associated with the normal shock wave, which
has been mentioned incidentally in several recent works (Chen, Chakravorty & Hung
1994; Stenholm & Jover 1997; Welsh 1997; Gribben et al. 1998; Frey & Hagemann
1998; Mat e et al. 2001). Somewhat more systematic numerical studies of the vortex
problem have been acomplished in the framework of the Euler equations by Goryainov
(2000), and of the quasi-gasdynamic equations by Graur et al. (2002b).
In the last decade, the quasi-gasdynamic (QGD) system of equations, and their
generalization (QGDR) for the breakdown of the translationalrotational equilibrium,
have been employed to calculate ow properties of several gasdynamic systems
(Elizarova et al. 1995, 1997; Elizarova & Chirokov 1999; Elizarova, Graur & Lengrand
Author to whom correspondence should be addressed: emsalvador@iem.cfmac.csic.es
240 I. A. Graur and others
2001; Graur, Elizarova & Lengrand 1997, 1999; Graur et al. 2002b; Lengrand et al.
1995). These equations constitute a promising gasdynamic approach beyond the
NavierStokes equations. However, the lack of quantitative experimental data to
compare with has left open a number of relevant questions about the limits of
applicability of the QGD approach in highly demanding computational problems like
the two-dimensional shock waves. Among them, a major one is to what extent several
dicult ow elds can be calculated accurately with the same method.
In this paper we present a comprehensive numerical and experimental study of
the two-dimensional problem of the shock wave and associated vortex formation in
axisymmetric jets of N
2
. The experimental part concerns the properties of ve normal
shock waves in axisymmetric expansions of N
2
generated under dierent stagnation-
to-background pressure ratios. These ratios were chosen to produce normal shock
waves diering considerably in their properties, spanning a range of Mach numbers
7.7 <M <15.3, reaching maximum values of local Knudsen numbers 0.33 <Kn <0.59
within the shock wave. The absolute density proles and the rotational temperature
proles of these ve shock waves have been measured with unprecedented accuracy
and spatial resolution by means of high-sensitivity Raman spectroscopy using the
miniature jet diagnostic facility at the Instituto de Estructura de la Materia (Montero
et al. 2000, Ramos et al. 2000).
The numerical interpretation is based on the QGD approach, and on its
QGDR generalization for the breakdown of the translationalrotational equilibrium,
particularly severe at the shock wave. The experimental material, rich in density
temperature features, provides a rm basis to test the capability of QGD and QGDR
modelling of shock waves which are in the continuum limit by virtue of their large
local Knudsen numbers. In particular, the present ow elds are characterized by the
following diculties: (a) a wide range of local Knudsen numbers, with experimental
values in the range Kn <0.59; (b) gasdynamic quantities (pressure, temperature, ow
velocity) varying by several orders of magnitude, with strong local gradients; and
(c) high rarefaction, and severe breakdown of rotationaltranslational equilibrium
in some local regions. Such high values of the local Knudsen number imply a
limit to the continuum models associated with the breakdown of the Maxwellian
distribution, while the breakdown of translationalrotational equilibrium poses the
problem of the rotational distribution function. In QGD and QGDR approaches this
distribution function is based on a continuous distribution of rotational energy, while
the molecular rotational quantum energies are of discrete nature, markedly departing
from a continuum at the low local temperatures reached at the onset of the present
shock waves, on the order of 10 K.
Testing dierent computational variants under these limit conditions, with emphasis
on the topics enumerated above, is the main target of the present work. Another
question that has received our attention is the formation of vortices beyond the
shock waves, akin to those described in the recent literature. The merits of the
QGDR generalization, compared with the plain QGD approach, and the use of
multiprocessor systems in connection with very ne computational grids are also
treated in some detail.
The paper is structured as follows. In 2, the theoretical background of the QGD
and QGDR equations is presented, with emphasis in the dierences with the Navier
Stokes equations. The experimental aspects of the shock waves studied are described
in 3. Computational and numerical treatment of the QGD and QGDR equations
is given in 4. A discussion of the density, temperature, and velocity proles of
the ve shock waves, calculated with dierent approximations, is presented in 5,
Shock waves in expanding ows 241
including comparison with experiment where possible. The numerical results and the
vortex problem are discussed in 6. Finally, the conclusions are summarized in 7.
Appendices A and B include details of the QGD equations adapted to axial symmetry,
and its vectorial representation for numerical calculations.
2. Theory
2.1. The QGD and QGDR equations
The quasi-gasdynamic (QGD) equations, originally developed by Elizarova &
Chetverushkin (1984, 1988) on the basis of a kinetical model of the distribution
function, are reputed to produce robust numerical algorithms suited for the calculation
of viscous supersonic ows. For the stationary case, the NavierStokes (NS) equations
are the asymptotic limit of the QGD equations when the Maxwellian relaxation time
tends to zero. This property can easily be deduced from the formulation of the
QGD equations presented below.
In the domain of small Knudsen numbers where the NS equations are accepted
to be valid, the QGD equations do not distort the NS solution but just stabilize
the numerical algorithm. In this case, QGD, NS, and direct simulation Monte Carlo
(DSMC) results, tend to coincide as has been shown by Elizarova et al. (1995, 1997),
and are in good general agreement with the experimental results (Graur et al. 2003).
For larger Knudsen numbers, QGD results have proved to be superior to NS results,
compared with experiment in microchanneles (Elizarova & Sheretov 2002). QGD
calculations, though less accurate than DSMC calculations, are considerably cheaper
from the computational point of view, and appear to be better suited for problems
where the gasdynamic properties span a range of several orders of magnitude, like in
the jets and shock waves studied in the present work.
Gasdynamic structures may be described by a system of three dierential equations
accounting for
conservation of mass (continuity equation)

t
+
i
J
i
=0, (1)
momentum
(u
k
)
t
+
i
J
i
u
k
+
k
p =
i

ik
, (2)
and total energy
E
t
+
i
J
i

(E + p) +
i
q
i
=
i
(
ik
u
k
), (3)
where the macroscopic ow quantities are (density), u
i
(velocity coordinates), p
(pressure), and E (total energy per unit volume).
Dierent choices for the mass ux vector J
i
, the shear-stress tensor
ik
, and the
heat ux vector q
i
, lead either to the NS equations, or to the QGD equations used
below (Sheretov 1997, 2000). The NavierStokes equations are derived from
J
i
=J
i
NS
=u
i
, (4)

ik
=
ik
NS
=[
k
u
i
+
i
u
k
(2/3)g
ik

j
u
j
] + g
ik

j
u
j
, (5)
q
i
=q
i
NS
=
i
T, (6)
242 I. A. Graur and others
where is the viscosity, g
ik
is the metric tensor, is the bulk viscosity, and =c
p
/Pr
is the heat conductivity; Pr is Prandtl number, and =c
p
/c
v
the ratio of heat
capacities; has been taken as =(5/3 ) according to Elizarova & Chirokov
(1999). The gasdynamic variables , u
i
, and p involved are instantaneous space-
averaged quantities. In contrast to the NavierStokes equations, if the gasdynamic
quantities , u
i
, and p are dened by means of timespace averaging (instead of space
averaging), the system (1)(3) can be closed in the QGD approach by
J
i
=J
i
NS
+ J
i
D
, J
i
D
=[
j
(u
i
u
j
) +
i
p], (7)

ik
=
ik
NS
+
ik
D
,
ik
D
=u
i
[u
j

j
u
k
+
k
p] + g
ik
[u
j

j
p + p
j
u
j
], (8)
q
i
=q
i
NS
+ q
i
D
, q
i
D
=u
i
[u
j

j
+ pu
j

j
(1/)], (9)
where =/p is the averaging time, chosen to be equal to the Maxwellian relaxation
time, and =p/(( 1)); ( )
D
stands for the dissipative terms additional to the
NavierStokes contributions referred to as ( )
NS
(Sheretov 1997, 2000).
The QGDR equations were developed by Elizarova & Chirokov (1999) as a
generalization of the QGD equations. The QGDR equations for a gas with three
translational and two rotational degrees of freedom, with =7/5, are intended
to account for the non-equilibrium T
tr
=T
rot
between translational and rotational
temperatures. The QGDR equations can be written like the QGD ones, with equations
(1) and (2) retaining their form, while in (7) and (8) p is replaced by p
tr
, and
by
tr
=/p
tr
, with the viscosity =(T
tr
) only depending on the translational
temperature. In turn, the energy equation (3) is split in the QGDR generalization into
one equation for the translational energy E
tr
per unit volume,
E
tr
t
+
i
J
i

(E
tr
+ p
tr
) +
i
q
i
tr
=
i
(
ik
u
k
) + S
tr
, (10)
and another equation for the rotational energy E
rot
per unit volume,
E
rot
t
+
i
u
i
E
rot
+
i
q
i
rot
=
i

j
u
i
u
j
E
rot
+
i

p
rot

i
p
tr
+ S
rot
, (11)
with
E
tr
=
(u
i
)
2
2
+
p
tr
1
, E
rot
=p
rot
, (12)
and S
tr
and S
rot
the energy exchange terms dened below.
The total heat ux q
i
is also split into the partial contributions
q
i
tr
=
1
Pr
5
2

i
p
tr

u
i
[u
j

j
+ p
tr
u
j

j
(1/)], (13)
where =p
tr
/(( 1)), responsible for the heat conduction due to the gradient of
translational temperature T
tr
, and
q
i
rot
=
1
Pr

i
p
rot

, (14)
for the heat conduction due to the gradient of rotational temperature T
rot
. Note
that in the QGDR model both heat ux vectors, q
tr
and q
rot
in (13) and (14), are
proportional to (T
tr
). The viscosity has been treated within the variable hard sphere
Shock waves in expanding ows 243
(VHS) model, which leads to a thermal dependence
=
ref
_
T
T
ref
_

. (15)
The VHS molecular diameter of N
2
employed in the present calculations is d
ref
=
d(T
ref
) =4.1710
10
m, while
ref
=(T
ref
) =1.65610
5
Ns m
2
at T
ref
=273 K, and
=0.74 (Bird 1994).
The energy exchange terms
S
tr
=S
rot
=
3
5
rot
(p
rot
p
tr
), (16)
in (10) and (11) involve the rotational relaxation time
rot
. Here it has been estimated
as
rot
=Z
c
, where
c
=(7 2)(5 2)/30 is the mean collisional time, and
Z=
Z

1 +
_

3/2
/2
_
(T

/T
tr
)
1/2
+ ( +
2
/4)(T

/T
tr
)
, (17)
the so-called rotational collision number (Parker 1959); for N
2
, the parameters
Z

=23, and T

=91.5 K, have been taken from Bird (1994).


Average pressure p
av
and temperature T
av
derived from the QGDR generalization,
to be compared to the homologous QGD quantities p and T , are dened by
p
av
=(3 p
tr
+ 2 p
rot
)/5 = (R/M) T
av
, (18)
where R=8.31451 J K
1
mol
1
is the universal gas constant, and M the molar mass
of N
2
. At thermal equilibrium p
tr
=p
rot
=p
av
=p, resulting
E =E
tr
+ E
rot
=
u
2
i
2
+
p
1
, (19)
for the total energy per unit volume, the QGDR system becoming a single-temperature
QGD system with perfect-gas specic heat ratio =7/5 and Prandtl number
Pr =14/19.
The QGD equations in r, z coordinates adapted to the axial symmetry of the
present problem are given explicitly in Appendix A. In the present formulation the
NS-dissipative terms, ( )
NS
, (equations (A7), Appendix A), and the additional
dissipative terms, ( )
D
, (equations (A5), (A6), (A8), Appendix A) are written
separately. For stationary ows the terms in have asymptotic order of O(
2
) for
0 or, in the dimensionless form O(Kn
2
) (Sheretov 2000).
3. Experimental
Recent progress in instrumentation for high-sensitivity Raman spectroscopy,
combined with the design and implementation of suitable expansion chambers, has
provided a powerful diagnostic tool for the quantitative investigation of supersonic
jets and associated phenomena (Montero et al. 2000, 2002; Ramos et al. 2000; Mat e
et al. 2001).
Using the facility at the Instituto de Estructura de la Materia (CSIC), the density
and rotational temperature proles of ve axisymmetric expansions of N
2
and their
normal shock waves, including downstream wakes, have been investigated with high
accuracy and spatial resolution. These expansions were generated through a circular
nozzle of exit radius r
e
=0.1565 mm and internal length 1 mm, at room temperature
T
0
=295 K, and xed nominal stagnation pressure of 1 bar. Due to the miniature
244 I. A. Graur and others
size of the facility, this reference pressure was measured with a pressure transducer
connected with the nozzle prechamber by a long and narrow duct. Thus it is just an
orientative nominal pressure reference. The actual pressure in the nozzle prechamber,
P
0
=732 mbar (10%), was determined by spectroscopic measurement of the absolute
number density at dierent points of the expansion, some of them very close to the
nozzle, extrapolating the tted function to the nozzle origin located about one nozzle
radius inside the nozzle channel (Ramos 2001).
For the conditions mentioned above the expansions are free from any detectable
condensation. The ve shock waves, henceforth denoted by A, B, C, D, and E, were
formed at distances from the nozzle L=2.7, 6.1, 8.5, 11.6, and 15.2 mm by setting the
residual pressure in the expansion chamber to P

=4.2, 1.0, 0.5, 0.28, and 0.18 mbar,


respectively, by means of an inlet needle valve, always maintaining the same nozzle
prechamber pressure of P
0
=732 mbar.
Referred to the critical conditions at the nozzle exit (Mach number M =1) the
Reynolds number Re
e
=2r
e
e
u
e

e
/
e
of the investigated expansions is Re
e
3200. The
ow regime in the mixing layer at the beginning of the expansions can be characterized
by a Reynolds number referred to the distance L between nozzle and Mach disk,
dened as Re
L
=Re
e
/

P
0
/P

(Volchkov et al. 1973; Novopashin & Perepelkin 1989):


for Re
L
>10
4
the ow is turbulent, the range 10
3
<Re
L
<10
4
corresponds to the
laminarturbulent transition, for 10
2
<Re
L
<10
3
the ow is laminar, and in ows
with Re
L
<10
2
the eects of rarefaction inuence the gasdynamic structure. For
expansions A to E respectively Re
L
=243, 119, 84, 63, and 50, meaning that the
ow in expansions A and B is laminar, while in expansions C, D, and E there is
increasingly rareed ow where the gasdynamic structure may be expected to be
progressively inuenced.
With aid of the expression for the discharge coecient C
D
in terms of Re
e
(Benedict
1966; Sreekanth, Prasad & Prasad 1991) we obtain for the present experiments a
discharge coecient C
D
0.90. As shown below, this departure from ideal isentropic
conditions is consistent with a reduction of the nozzle radius from its nominal value
r
e
=0.1565 mm to an eective radius r
e
e
0.148 mm. These facts, indicative of the
departure from the ideal source conditions, may add some degree of quantitative
uncertainty to the comparison between numerical and experimental results, but are
not suciently important to aect seriously the general conclusions on the structure
and quantitative properties of the fairly dierent shock waves and wakes investigated
here.
As a guide for the forthcoming dicussion a QGD numerical schlieren picture of the
global structure of expansion B is depicted in gure 1(a) showing the representative
zones of the ow eld, which are common to the ve expansions A to E, but dier
quantitatively. Immediately downstream from the nozzle is the zone of silence, the
supersonic core where the velocity ow eld is nearly radial, with rapid decreasing
density and temperature, and increase of ow velocity approaching its asymptotic
(terminal) value, followed by the normal shock wave, of width , where the ow eld is
sharply deected outwards from the symmetry axis, with strong decreasing of the ow
velocity. The normal shock wave is characterized by a sharp increase of density and
temperature across it, and by a marked breakdown of the translationalrotational
thermal equilibrium. The nal section of the normal shock wave merges with a
rethermalized zone of slow subsonic ow velocity. According to some computational
results a toroidal trapped vortex, displaying negative axial ow velocity, is formed
there. This is indeed a controversial region because of the extremely slow ow velocity,
hardly amenable to unambiguous experimental conclusions. Beyond this region the
Shock waves in expanding ows 245
Figure 1. Expansion B: (a) QGD numerical schlieren picture of expansion B, variant s10,
(b) and the computational domain.
ow eld is nearly sonic and tends to collimate parallel to the symmetry axis. The zone
of silence, normal shock wave, and supposed vortex are conned by the barrel shock,
almost rethermalized and of comparatively high density, with slightly supersonic ow
velocity. Outside the barrel shock, close to it the ow is slow and probably turbulent,
while at distance large enough the gas may be considered at rest at the background
pressure p

. A thorough description of the general properties of free jets has been


given by Volchkov et al. (1973), and by Rebrov (1985), while a quantitative mapping
of a real free jet of CO
2
has been reported by Mat e et al. (2001).
The region experimentally investigated here in detail is the axial path of gure 1(a).
The two-dimensional QGD and QGDR calculations span a computational domain,
shown in gure 1(b), large enough to include the complete structure of the jet in all
ve cases A to E.
The experimental spatial resolution at the datapoints probed along the axis of
the expansion of gure 1 amounts about 10 m, equivalent to 0.0638 nozzle radius.
This value is far smaller than the experimentally determined widths of the shock
waves,
A
=0.606,
B
=2.556,
C
=4.920,
D
=6.912, and
E
=10.480 also expressed
in units of the nozzle radius r
e
, and allows the determination of truly local properties
even under strong gradient conditions. In terms of mean free path, more meaningful
from the physical point of view, the ve shock waves display a remarkably constant
thickness, with avearage value =(4.37 0.38)
1
; is dened here according to Bird
(1994), and
1
is the mean free path at the minimum density point upstream of each
shock wave (Ramos et al. 2000).
246 I. A. Graur and others
Computational Grid z-step r-step Domain
variant N
z
N
r
h
z
/r
e
h
r
/r
e
size (mm)
s1 140 92cv 1 0.1 21.91 15.65
s2 281 92cv 0.5 0.1 21.91 15.65
s3 561 92cv 0.25 0.1 21.91 15.65
s4 1401 1001u 0.1 0.1 21.91 15.65
s5 34 60cv 1 0.1 5.32 3.14
s6 137 201u 0.25 0.1 5.32 3.14
s7 171 201u 0.2 0.1 5.32 3.14
s8 340 201u 0.1 0.1 5.32 3.14
s9 76 74cv 1 0.1 11.73 7.84
s10 751 501u 0.1 0.1 11.73 7.84
Table 1. Computational variants used in the present work; u denotes uniform grids and cv
constant-variable grids.
The experimental density and rotational temperature proles of shock waves A, B,
C, D, and E, are discussed below, jointly with their QGD or QGDR calculated
counterparts under dierent computational conditions. More details about the
procedure for measuring of densities and rotational temperatures in the jet, as well
as other features of these shock waves have been reported by Ramos et al. (2000).
4. Numerical details
The computational domain of gure 1(b) is covered by a rectangular grid with radial
steps h
r
, and axial steps h
z
. Since the choice of grid and the size of the domain have a
determining inuence on specic features of the calculated ow elds, in particular on
the supposed vortex beyond the normal shock waves, several computational variants,
listed in table 1, have been tested.
Two sorts of grids were used, namely (i) uniform grids marked u in table 1 where
the step in the axial (z) and radial (r) directions remains constant, and (ii) constant-
variable grids marked cv in table 1 where the step size remains constant along the
axial direction, while for the radial direction it remains constant for r <r
e
, but for
r >r
e
the step increases progressively between adjacent cells with a stretching factor
1.05.
The numerical computation of the QGD and QGDR equations for expansions A
to E has been carried out according to a nite-dierence scheme where the spatial
derivatives are approximated with the centred derivatives (centred scheme). Under
high Mach numbers the numerical solution of the QGD and QGDR systems may
lead to oscillations in regions with strong gradients in the dynamical variables. To
overcome this numerical problem the relaxation time is replaced in the dissipative
terms, except in those with mixed spatial derivatives, by
=/p +
e
, with
e
=
h
c
, (20)
where c is the velocity of sound; the damping parameter , usually in the range
0.1 < <0.5, is given the smallest value that ensures the stability of the solution. This
makes the order of the space accuracy of the resulting numerical algorithm equal to
O(h). (See Appendix B for the details.)
Shock waves in expanding ows 247
M
e
1.01 Kn
e
1.09 10
4
T
e
249.2 K

e
=
e
/(2r
e
) 3.75 10
4
p
e
0.382 bar T

/T
e
200/249.2
n
e
1.1114 10
25
m
3
p

/p
e
(A) 0.0110

e
0.517 kg m
3
(B) 0.00262
u
e
325.0 ms
1
(C) 0.00131

e
15.47 10
6
s Pa (D) 0.000732

e
11.73 10
8
m (E) 0.000471
Table 2. Nozzle exit quantities of the N
2
jets employed in the QGD and QGDR numerical
calculations; Kn
e
=(
e
/n
e
)(dn/dz)
e
.
The system of nite-dierence equations associated with the QGD and QGDR
equations is solved here by means of an explicit algorithm where the steady-state
solution is attained as the limit of a time-evolving process. The computation nishes
when the steady-state solution is reached according to the criterion
1
N
r
N
z

j+1

6, (21)
where the sum is over all computational nodes in the grid; is the density, j the
time-step index, N
r
, N
z
the number of nodes in the r- and z-directions respectively,
and

t the dimensionless time step dened as

t =(c
e
/
e
)t , where
e
is the mean
free path at the nozzle exit. Here t is chosen according to the stability criterion
t =0.005 min(h/c). Representative time steps and the number of iterations required
for a stationary solution in a QGDR calculation of shock waves A to E (variant s2)
are: 10
8
, N
iter
5 10
6
, and

t =2.40, 0.477, 0.240, 0.242, and 0.121, for shock


waves A, B, C, D, and E, respectively.
4.1. Parallel implementation
The numerical problem of shock wave calculations considered here is very time
consuming and requires a powerful computational system. We employed a cluster
computer system consisting of up to 300 processors with distributed memory equipped
either with Intel Pentium III, or with Alpha 21264 microprocessors. The Message
Passing Interface (MPI) standard was used for the organization of interprocessor data
exchange. The parallel code is geometrical-parallelism based, according to the domain
decomposition technique. This means that the whole computational domain is divided
into subdomains in the z-direction, with each processor providing the calculations in
its own subdomain. The number of subdomains is equal to the number of processors
used. Eciency estimations show that the implemented numerical algorithm, explicit
in time and homogeneous in space allows the ecient use of the cluster multiprocessor
systems (Graur et al. 2002a).
4.2. Flow and boundary conditions
The QGD and QGDR numerical calculations were carried out under ow conditions
suited for comparison with the experimental results. The nozzle exit quantities, some of
which are required for the calculation, are reported in table 2. They have been obtained
from the source conditions of the experiment assuming an isentropic approximation
with Ma
e
=u
e
/a
e
=1.01 at the exit of the nozzle. The ratio of background to nozzle
exit pressures of the ve shock waves investigated are marked (A) to (E) in table 2.
248 I. A. Graur and others
A representative two-dimensional computational domain is shown in gure 1. The
PQ, QR, RS, and SN, boundary conditions for the QGD and QGDR equations are
as follows:
PQ (nozzle wall boundary)
u
z
=0,
u
r
z
=0,
p
z
=0,
T
z
=0, (22)
QR (radial undisturbed boundary)
u
z
r
=0, u
r
=0, p =p

, T =T

, (23)
RS (downstream boundary)
u
z
z
=0,
u
r
z
=0,
p
z
=0,
T
z
=0, (24)
SN (symmetry axis boundary)
u
z
r
=0, u
r
=0,
p
r
=0,
T
r
=0. (25)
4.3. Schlieren pictures
The global representation of the two-dimensional problem discussed next has been
generated by means of a numerical schlieren picture. According to Liepmann &
Roshko (1957) either /r or /z can be visualized in experimental schlieren
pictures depending on the knife position, vertical or horizontal, respectively. For
numerical visualization both possibilities can be used but, in accordance with our
experience, the quantity best suited for a faithful picture is based on the absolute
value of the gradient
|| =

z
_
2
+
_

r
_
2
. (26)
In order to expose even weak density non-uniformities, a nonlinear scale has been
utilized as proposed by Quirk (1994, 1998). A schlieren picture corresponding to the
ow eld of expansion B is shown in gure 1. It depicts the quantity
S(z, r) =C exp
_
K
|| ||
min
||
max
||
min
_
, (27)
where the subscripts min and max denote the minimum and maximum values of the
density gradient over the whole ow eld; C and K are two tunable parameters.
The parameter C determines the shade of grey that corresponds to the zero gradient,
while K governs the amplication of small gradients. We have used C =0.8, and K
between 10 and 15.
5. Density, temperature, and velocity proles of shock waves A, B, C, D, and E
The density, rotational temperature, and velocity, proles of axisymmetric
supersonic expansions and associated shock waves are characterized by a number
of reference points, depicted schematically in gure 2 for a generic expansion. These
points are located along the symmetry axis of the expansion.
Reference point 1, of abscissa z
1
, in gure 2(a) corresponds to the minimum
of density, n
1
, in the zone of silence of the expansion. Point 2 corresponds to a
Shock waves in expanding ows 249
Figure 2. Reference points in a generic axisymmetric supersonic expansion; (a) densities,
(b) rotational temperatures, and (c) velocities.
250 I. A. Graur and others
A B C D E
z
1
QGDR 2.50 4.70 6.50 8.53 10.64
Exptl 2.55 5.3 7.5 10.0 13.0
n
1
QGDR 21.46 4.84 2.27 1.18 0.70
Exptl 30.6 6.3 3.3 1.8 0.93
L (Mach disk) QGD
a
2.58 4.93 7.75 9.16 13.79
QGDR
a
2.74 5.24 7.90 10.80 13.93
Exptl
a
2.7 6.1 8.5 11.6 15.25
z
min
Ashk.
b
2.62 5.37 7.59 10.14 12.65

max
QGDR 0.16 0.74 1.57 2.96 4.86
z(

max
) QGDR 2.58 4.85 6.81 8.92 11.19
Kn
max
Exptl 0.33 0.35 0.42 0.54 0.59
z(Kn
max
) Exptl 2.68 6.0 8.3 11.3 14.8
Kn
max
QGDR 0.62 0.87 1.06 1.40 1.79
z(Kn
max
) QGDR 2.66 5.09 7.28 9.70 12.36
a
from maximum density gradient,
b
from equation (28) according to Ashkenas & Sherman (1964),
employing the eective nozzle radius r
e
e
=0.148 mm.
Table 3. Numerical results on shock waves A, B, C, D, and E for computational variant s2,
and comparison with experiment; z and L are distances downstream from the nozzle, in mm;
n
1
is the minimum absolute number density on the axis, in units of 10
21
m
3
;

=/(2r
e
) is
the normalized mean free path, and Kn =(/n)(dn/dz) the local Knudsen number.
discontinuity in the slope of the density. To experimental accuracy, this discontinuity
coincides with the highest rethermalization of the rotational temperature, about
20% above the nozzle exit temperature, T
e
, as shown in gure 2(b). Along the
expansion the rst experimental evidence of the onset of the normal shock wave is the
rethermalization, which starts upstream from point 1, of minimum density, shown in
gure 2(b). This agrees with the predictions of Rebrov & Chekmarev (1971) for a
spherically expanding ow, a good model for the paraxial region of the present ex-
pansions up to point 2. Point 2 may be considered the end of the actual shock wave.
Between points 1 and 2, the abscissa L of largest slope in the density prole of
gure 2(a) coincides with the sonic condition M =1. Customarily this point is referred
to as the location of the Mach disk.
According to Ashkenas & Sherman (1964) the abscissa of minimum pressure
upstream from the shock wave is given by
z
min
=1.34r
e
(p
0
/p

)
1/2
, (28)
where r
e
is the radius of the nozzle, p
0
the stagnation pressure in the nozzle
prechamber, and p

the background pressure in the expansion chamber. The abscissa


of point 1, z
1
, its absolute number density, n
1
, the location L of the Mach disk,
and other relevant ow parameters of cases A to E are given in table 3, from the
experiment, and computational variant s2. It should be noted that the abscissa of
lowest pressure, z
min
, from (28), is indeed very close to the experimental point 1 of
lowest number density, but progressively departs from the true position of the Mach
disk on going from shock A to shock E.
For all grids of table 1 QGDR results are somewhat closer to the experimental
values. QGD results slightly underestimate the position of the Mach disk compared
to QGDR results.
Shock waves in expanding ows 251
The density jump across the shock, between points 1 and 2 of gure 2(a), is found
empirically to obey the relation
n
2
n
1
=
_
z
1
z
2
_
2
( + 1)M
2
2 + ( 1)M
2
, (29)
where z
1
and z
2
are the axial distances from the nozzle to points 1, and 2, respectively,
and M is the Mach number at the onset of the normal shock. The experimental
values M7.7, 10.6, 12.2, 13.6, and 15.3, at the onset of shocks A, B, C, D, and E,
respectively, have been estimated from the distance between the nozzle and the point
of lowest rotational temperature according to the empirical parameterization of M as
a function of z and proposed by Miller (1988). In the limit of innitesimal width
(z
2
=z
1
), equation (29) leads to the RankineHugoniot density jump across an ideal
monodimensional shock wave. For increasingly larger widths of the real shock wave,
i.e. for increasing mean free path, the discontinuity in the gradient of density at point
2 tends to vanish. In shock wave E it is no longer recognizable, but its location can
be inferred approximately from the rethermalization of the rotational temperature.
Between points 2 and 3 the density gradient is positive, but smaller than at the
shock, and at point 3 the density reaches a local maximum. Downstream from point
3 mild secondary expansions are observed, with a quasi-periodical structure of density
maxima (points 3), and minima (points 4). As shown in gures 2(a) and 2(b), these
density maxima and minima are correlated with temperature maxima and minima,
but are anticorrelated with those of ow velocity (gure 2c).
The QGD and QGDR numerical simulation of these features shows an encouraging
qualitative agreement with experiment for all ve shock waves investigated. To what
extent the simulation reaches a quantitative agreement very much depends on the
grid employed, on the size of computational domain, on the sort of approximation,
QGD versus QGDR, on the damping parameter dened previously (equation (20)),
and on the rarefaction at the onset of the shock. The inuence of these factors is
discussed next.
5.1. Shock wave A
The convergence of the QGD numerical solution with grid renement is summarized
in gure 3 for variants s1 to s4 (see table 1). Density proles improve beyond point 2
for ner grids, as shown in gure 3(a), and in more detail in gure 4, calculated with
the nest grid in a reduced computational domain. Altough variant s8 is still unable
to reproduce the slope beyond point 2, the density jump between points 1 and 2 is
predicted very accurately, as well as the slope between points 1 and 2. The Mach disk
location L is not very sensitive to grid nor to domain size. For the variants of table 1
this position is in the range 2.504 6L62.676 mm, on average 5% smaller than the
experimental value L(A) =2.7 mm.
Calculated temperature proles shown in gure 3(b) slightly improve for ner grids
only in the region of lowest temperatures prior to the shock wave. Otherwise, the
calculated thermal prole of shock wave A is nearly insensitive to grid choice, at least
at the spatial resolution of the present experiments. In general QGD calculations
predict an average thermal prole which is in fairly good agreement with experiment,
as shown in gure 3(b). In particular the thermal overshoot at point 2 is well
reproduced. The sequence of maxima (points 3) and minima (points 4) for densities
and temperatures shown in gure 3(b) is also well reproduced regardless of grid.
QGD calculated axial velocities are depicted in gure 3(c). A recirculation vortex
characterized by negative axial velocities, located immediately downstream from
252 I. A. Graur and others
Figure 3. Shock wave A: experimental and QGD calculated axial proles of density
(a), temperature (b), and velocity (c), as a function of the grid; variants s1, s2, s3, and s4.
point 2, is obtained for cv grids in variants s1, s2, and s3, but not for the uniform
nest grid in variant s4. Though the QGD description of the density prole between
point 2 and the rst point 3 is not good (gradient too large) it may be expected
Shock waves in expanding ows 253
Figure 4. Shock wave A, detail: experimental and QGD calculated axial proles of density
(a), and temperature (b); variants s5 and s8.
that variant s4, yielding the closest gradient to the experiment, would also predict
the velocity eld best between 2 and 3. Indeed, in variant s4 the vortex disappears,
suggesting that the vortex resulting in shock wave A for grids s1, s2, and s3, is a
computational artifact with no physical reality.
The vortex also tends to vanish when the damping parameter is increased from
=0.1 to 0.5, but this is probably caused by a smoothing of the calculated gradients
of all quantities, including those of the velocity.
The dierences between the QGD and the QGDR generalization are minor for
shock wave A, as shown in gure 5 for variant s2. The QGDR prediction of Mach
disk position, L=2.74 mm, is closer to the experimental value L=2.7 mm, but the
sequence of maxima and minima, points 3 and 4, is not as good as for QGD
calculation. As far as the presumed vortex is concerned, QGD and QGDR numerical
results are similar.
5.2. Shock wave B
The convergence of QGD density proles with grid renement is shown in gure 6.
In particular the density gradient around point 2 is well reproduced by variant s3.
The thermal gradient is, however, less sensitive the to grid choice, as was the case
254 I. A. Graur and others
Figure 5. Shock wave A: QGD and QGDR calculated axial proles of density (a), and
rotational temperature (b); variant s2.
with shock wave A. Only a minor improvement is obtained with grid renement. The
position of the Mach disk is underestimated by about 20%, although the trend is
towards an improvement with grid renement. As was the case with shock wave A,
shock wave B also displays a recirculation vortex for variants s1, s2, and s3.
The eect of using a uniform grid in a smaller computational domain, variant s10,
is shown in gure 7. The density prole improves substantially between points 2 and
3, the region corresponding to the presumed vortex, but there is little change in the
thermal prole. The pressure gradient appears be well reproduced by the calculation
variant s10, so the small vortex in gure 7(c) between z 6 and 7 mm, with reversed
ow velocity u
z
0.1 u
ze
is obtained as a persistent numerical solution.
The QGDR generalization, even at the computational level of the coarse grid
of variant s2, yields a substantial improvement compared to QGD density and
temperature proles, as shown in gure 8. The position of the Mach disk improves,
and QGDR density maxima and minima, points 3 and 4, are very well reproduced,
as are rotational temperatures above 150 K. The thermal onset of the shock wave,
below 100 K, is, however, poorly reproduced by any computational variant. The initial
section of the calculated thermal prole at the shock wave is far smoother than the
Shock waves in expanding ows 255
Figure 6. Shock wave B: experimental and QGD calculated axial proles of density
(a), temperature (b), and velocity (c), as a function of the grid; variants s1, s2, and s3.
experimental one. This mismatch is already evident in shock wave A, and worsens
considerably the more rareed is the ow at the point 1 of the shock wave. QGDR
and QGD numerical behaviours of the vortex are similar.
256 I. A. Graur and others
Figure 7. Shock wave B, detail: QGD calculated axial proles of density (a), temperature
(b), and velocity (c); variants s9 and s10.
5.3. Shock waves C, D, and E
Numerical and experimental results concerning shock waves C, D, and E, are depicted
in gures 9 to 13. For these shock waves, with Re
L
<100, the inuence of the grid on
Shock waves in expanding ows 257
Figure 8. Shock wave B: QGD and QGDR calculated axial proles of density
(a), rotational temperature (b), and velocity (c); variant s2.
the QGD numerical results is of decreasing importance, as can be seen from gure 9
(shock wave C), and gure 11 (shock wave D). This is not surprising since the density
and temperature gradients tend to be smoother the lower the density at the onset of
the shock wave.
258 I. A. Graur and others
Figure 9. Shock wave C: QGD calculated axial proles of density (a), temperature (b), and
velocity (c), as a function of the grid; variants s2 and s3.
QGDR proles are substantially closer to experiment than QGD proles as shown
in gure 10 (shock wave C), gure 12 (shock wave D), and gure 13 (shock wave E).
This proves the superiority of QGDR over the plain QGD approach in these systems,
Shock waves in expanding ows 259
Figure 10. Shock wave C: QGD (primed numbers) and QGDR calculated axial proles of
density (a), and rotational temperature (b); variant s2.
or regions, of the ow eld where rotationaltranslational relaxation eects are far
from neglegible.
The numerical recirculation vortex appears for variants s2 and s3 in shocks C and
D, but for shock D it disappears for variant s10. For shock E no vortex is detectable
in the calculated ow eld of velocities, but is obtained for the mass ux vector
eld J
i
.
6. Discussion
The absolute number density along expansions A to E varies from n
e
=1.1
10
25
m
3
, at the nozzle exit, up to a minimum value n
1
at point 1, three to four
orders of magnitude smaller than n
e
. A remarkable merit of the QGDR approach
is its capability of accounting for this wide range of absolute number densities with
fractional deviation
=

n
QGDR
1
n
exptl
1

n
e
n
1
, (30)
well below 10
3
for all ve expansions, according to the values given in table 3. In
gures 3 to 13 it is shown how the normalized density proles of shock waves A to
260 I. A. Graur and others
Figure 11. Shock wave D: experimental and QGD calculated axial proles of density
(a), and temperature (b), as a function of the grid; variants s1, s2, s3, and s4.
E are reproduced by the various computational variants described in the previous
sections.
The systems studied here are associated with increasingly rareed ows where the
maximum value, Kn
max
, of the local Knudsen number dened in the caption of table 3,
is reached in the initial section of the shock wave; Kn
max
progresively increases from
shock wave A to shock wave E, as shown in table 3. These high values of Kn
max
are due to the large mean free path at the onset of the shock wave (, inversely
proportional to the number density, increases here by three to four orders of magnitude
from nozzle to shock wave), divided by a small density, and multiplied by the large
density gradient within the shock wave. This large variation of local Knudsen number
between nozzle and shock wave, together with its intrinsic high value inside the shock
wave, suggests that the present problem is in the limit of applicability of continuum
models like QGD and QGDR.
It may be argued however whether the poor description of the QGDR rotational
temperature proles in the low-temperature section prior to the softer shock waves
(see gures 10b, 12b, and 13b) is caused by an inaccurate estimate of the viscosity
for T <100 K. In order to check this point shock wave E, exhibiting the largest
local Knudsen number, has been recalculated in the QGDR approach using a
Shock waves in expanding ows 261
Figure 12. Shock wave D: QGD and QGDR calculated axial proles of density (a), and
rotational temperature (b); variant s2.
Sutherland-type viscosity law (Hirschfelder, Curtiss & Bird 1954) for nitrogen, where
=1.374 10
6
T
3/2
T + 100
for T >100 K, (31)
=1.374 10
6
T
2

100
for T <100 K. (32)
This viscosity law does indeed improve the lowest value of the rotational temperature
in case E by 30%, but does not improve signicantly the too smooth nature of
the thermal onset of the shock wave. So we conclude that the too smooth calculated
thermal gradient at the onset of the shock waves is an intrinsic limitation of the
models based in the hypothesis of a continuum.
This limitation may be aggravated under conditions of severe breakdown of
rotationaltranslational equilibrium, as is the case within the present shock waves.
This is an unavoidable limitation of the QGD and QGDR models, and in general
of continuum models, where the rotational distribution function is based on the
hypothesis of continuous rotational energy, ignoring the discrete nature of rotational
quantum levels. The very low temperatures at the beginning of the present shock
waves preclude any reasonable averaging of the rotational distribution function.
262 I. A. Graur and others
Figure 13. Shock wave E: experimental, QGD and QGDR calculated axial proles of
density (a), and rotational temperature (b); variant s2.
The above problems may in part be overcome by the direct simulation Monte Carlo
method (DSMC). But in the particular problem treated here it brings additional
computational diculties due to the large dierences of geometrical features in cases
A to E, with very strong density and pressure gradients in the ow. This has been
shown by Teshima & Usami (2001), who had to employ dierent size cells and
dierent time-step schemes to treat a similar problem.
In spite of the specic discrepancies mentioned above the, overall agreement between
numerical and experimental proles of density and rotational temperature confer on
the present results, specially the QGDR ones, a reasonable degree of credibility.
A striking feature of the calculated ow elds is the steady recirculation vortex,
trapped immediately downstream from the shock wave, which appears as a stable
numerical solution when using coarse grids (Graur et al. 2002b). These sorts of
solutions are summarized in gure 14 showing an overview of the velocity eld. There
the nature of reference point 2 can be related to sudden changes in the orientation
of the velocity vector. Similar vortices have been described before in the literature,
but no experimental proof of their physical reality has been reported, as far as we
are aware. In the light of the present experimental and numerical results it is not
possible to conrm, or to deny, whether they are spurious numerical solutions or not.
Shock waves in expanding ows 263
Figure 14. QGDR ow elds of N
2
expansions A, B, C, D, and E, calculated with variant
s2; 2

indicates the calculated position of reference point 2; D=2r


e
is the nozzle diameter.
In the experiment, the very slow velocity beyond the shock wave, with or without
recirculation, is too low to be detectable with the spectroscopic technique employed.
The present numerical solutions indicate, however, the tendency of the vortex to
vanish, in most cases but not always, when ner computational grids are used. Even
with the nest grid compatible with our computational hardware resources, the vortex
of shock wave B remains as a stable numerical solution. Only a substantial increase
of these resources can in principle remove this ambiguity by employing still ner
grids. In any case it is worth noticing the extreme computational sensibility of the
264 I. A. Graur and others
velocity ow eld that can be inferred from gure 14 in the vortex region. There the
orientation and magnitude of the velocity vector appears as an extremely sensitive test
for gasdynamical models and computational variants. Note however that the present
approach is restricted by rz geometry, but the actual ow might be fully three-
dimensional, a possibility that should not be excluded a priori. The recirculation ow
behind the Mach disk might indeed be more complicated than a simple stationary
vortex with axial symmetry. Since in the present problem Re
e
3200, at the nozzle
exit the ow might be spiraliform, or might even have a more complicated topology,
as is known to happen in ows behind obstacles, even for rather small Re (Shevelev
1986; Perry & Chong 1987; Mason & Sykes 1979; Shevelev & Klekovkin 2003).
7. Conclusions
The present work shows to what extent a gasdynamic problem with a wide range
of local Knudsen numbers can be treated within the framework of the QGD/QGDR
approach. The following conclusions can be drawn:
(i) The QGD equations, and specially their QGDR generalization, provide a
reasonable description of axial densities, and a somewhat less accurate description
of temperatures, in two-dimensional ow elds of supersonically expanding N
2
, with
measured local Knudsen number up to Kn <0.6, approximately. These ow elds
include normal shock waves in the range of Mach numbers 7.7 <M <15.3.
(ii) The less satisfactory feature of the QGD approach is the description of
the thermal evolution at the onset of the shock waves, the most rareed region
of the ow eld. This numerical description of temperatures is moderately good
in regions with experimental local Knudsen number Kn 60.3. However, wherever
Kn 0.6 it worsens considerably. The QGDR generalization overcomes in part this
limitation but still predicts a too smooth prole for the rotational temperature at
the onset of the shock waves. Since the translational distribution function is free
from quantum eects, the translational temperatures may be expected to be better
described by the QGDR approach than the rotational temperatures, specially at
the lower end of the thermal scale. Unfortunately, this conjecture cannot be proved
from the present experiment since translational temperature is not amenable to direct
measure.
(iii) The gradients of the gasdynamic quantities decrease markedly with increasing
rarefaction and mean free path, on going from shock wave A to shock wave E. The
larger the mean free path, as in shock wave E, the less determinant is the choice of
computational grid. On the other hand, shock wave A is particularly sensitive to the
density and type of computational grid.
(iv) Parallel implementation of the QGD and QGDR equations in this problem has
proved highly advantageous.
(v) The physical reality of the recirculation vortex formed immediately downstream
from the shock wave associated with axisymmetric hypersonic expansions cannot
be proved unambiguously from the present experimental data, nor from the QGD
or QGDR numerical results. In some cases the numerical vortex disappears when
dense enough grids are employed in the calculation. The vortex must so far remain
conjectural in our opinion.
(vi) Linear Raman spectroscopy proves to be an instrumental technique very well
suited for the quantitative study of normal shock waves. For shock waves spanning
a wide range of their gasdynamic quantities it can provide
Shock waves in expanding ows 265
(a) sampling at very high spatial resolution (few m)
(b) absolute densities (10%) within several orders of magnitude
(c) rotational temperatures (5%) nearly unlimited in range.
This work was supported by the Russian Foundation for Basic Research, grant
N 01-01-00061, and by the Spanish DGESIC (MEC), research project PB97-1203.
Thanks are due to the referees for the valuable comments towards the improvement
of this paper.
Appendix A. The r, z-formulation
The present QGD and QGDR equations in the r, z formulation dier from the
version given by Mat e et al. (2001). The NS dissipative terms, and the additional
dissipative terms, are written here separately. For brevity we just report the r, z-
formulation for the QGD system. In the case of r, z geometry equations (1)(3) take
the form

t
+
1
r

r
(rJ
r
) +

z
J
z
=0, (A1)
(u
r
)
t
+
1
r

r
(rJ
r
u
r
) +

z
(J
z
u
r
) +
p
r
=
1
r

r
(r
rr
) +

z

zr

r
, (A2)
(u
z
)
t
+
1
r

r
(rJ
r
u
z
) +

z
(J
z
u
z
+ p) =
1
r

r
(r
rz
) +

z

zz
, (A3)
E
t
+
1
r

r
rJ
r
(E + p)

+

z
J
z
(E + p)

+
1
r

r
(rq
r
) +

z
q
z
=
1
r

r
r(
rr
u
r
+
rz
u
z
) +

z
(
zr
u
r
+
zz
u
z
), (A4)
where
E =
u
2
r
+ u
2
z
2
+
p
1
.
The expressions for the mass ux vector, the heat ux vector, and the shear stress
tensor are:
J
r
NS
=u
r
, J
r
D
=
_
1
r

r
_
ru
2
r
_
+
p
r
+

z
(u
r
u
z
)
_
,
J
z
NS
=u
z
, J
z
D
=
_
1
r

r
(ru
r
u
z
) +

z
_
u
2
z
_
+
p
z
_
,
_

_
(A5)
q
r
NS
=

1
1
Pr

r
_
p

_
, q
z
NS
=

1
1
Pr

z
_
p

_
,
q
r
D
=
_
u
r
1
_
u
r

r
_
p

_
+ u
z

z
_
p

__
+ u
r
p
_
u
r

r
_
1

_
+ u
z

z
_
1

___
,
q
z
D
=
_
u
z
1
_
u
r

r
_
p

_
+ u
z

z
_
p

__
+ u
z
p
_
u
r

r
_
1

_
+ u
z

z
_
1

___
,
_

_
(A6)
266 I. A. Graur and others

rr
NS
=2
u
r
r
( 1) u,

zr
NS
=
rz
NS
=
_
u
r
z
+
u
z
r
_
,

zz
NS
=2
u
z
z
( 1) u,

NS
=2
u
r
r
( 1) u.
_

_
(A7)

rr
D
=
_
u
2
r
u
r
r
+ u
r
u
z
u
r
z
+ 2u
r
p
r
+ u
z
p
z
+ p u
_
,

rz
D
=
_
u
2
r
u
z
r
+ u
r
u
z
u
z
z
+ u
r
p
z
_
,

zr
D
=
_
u
r
u
z
u
r
r
+ u
2
z
u
r
z
+ u
z
p
r
_
,

zz
D
=
_
u
r
u
z
u
z
r
+ u
2
z
u
z
z
+ 2u
z
p
z
+ u
r
p
r
+ p u
_
,

D
=
_
u
r
p
r
+ u
z
p
z
+ p u
_
,
_

_
(A8)
u =
1
r
(ru
r
)
r
+
u
z
z
.
It should be mentioned that in the r, z formulation the ow quantities are uniform
on azimutal angle , with vanishing derivatives on it. However, the stress tensor
component

is independent of , and is non-zero even in the r, z formulation.


Appendix B. Numerical realization
For the numerical realization the system (A1) (A4) is rewritten in the vector-form:
U
t
+
1
r
(r F)
r
+
F
1
r
+
1
r
(r F
D
)
r
+
E
z
+
E
D
z
=
1
r
(rW)
r
+
1
r
(rW
D
)
r
+
V
z
+
V
D
z

G
r

G
D
r
, (B1)
where
U =
_
_
_
_
_

u
r
u
z
E
_
_
_
_
_
, F =
_
_
_
_
_
J
r
NS
u
r
J
r
NS
u
z
J
r
NS
J
r
NS
(E + p)/
_
_
_
_
_
, E =
_
_
_
_
_
J
z
NS
u
r
J
z
NS
u
z
J
z
NS
+ p
J
z
NS
(E + p)/
_
_
_
_
_
, F
1
=
_
_
_
_
_
0
p
0
0
_
_
_
_
_
,
(B2)
F
D
=
_
_
_
_
_
J
r
D
u
r
J
r
D
u
z
J
r
D
J
r
D
(E + p)/
_
_
_
_
_
, E
D
=
_
_
_
_
_
J
z
D
u
r
J
z
D
u
z
J
z
D
J
z
D
(E + p)/
_
_
_
_
_
, G=
_
_
_
_
_
0

NS
0
0
_
_
_
_
_
, G
D
=
_
_
_
_
_
0

D
0
0
_
_
_
_
_
,
(B3)
Shock waves in expanding ows 267
W =
_
_
_
_
_
0

rr
NS

rz
NS
u
r

rr
NS
+ u
z

rz
NS
q
r
NS
_
_
_
_
_
, W
D
=
_
_
_
_
_
0

rr
D

rz
D
u
r

rr
D
+ u
z

rz
D
q
r
D
_
_
_
_
_
, (B4)
V =
_
_
_
_
_
0

zr
NS

zz
NS
u
r

zr
NS
+ u
z

zz
NS
q
z
NS
_
_
_
_
_
, V
D
=
_
_
_
_
_
0

zr
D

zz
D
u
r

zr
D
+ u
z

zz
D
q
z
D
_
_
_
_
_
. (B5)
As mentioned in 4, in order to stabilize the numerical algorithm the relaxation time
is replaced by +
e
in several dissipative terms of the QGD and QGDR systems.
In particular, in the QGD system this procedure is employed for the J
D
and q
D
vectors, and for the
D
and
NS
tensors as follows: the replacement is made only in
the terms of the system (A1) (A4) obtained by taking the derivatives of J
D
, q
D
,

D
, and
NS
, that would include second derivatives in the r and z coordinates. The
corresponding terms for the J
D
and q
D
vectors, and for the
D
tensor, are

z
( +
e
)
f
z
,

z
( +
e
)g
g
z
,
1
r

r
( +
e
)

r
(rf ),
1
r

r
( +
e
)g

r
(rg),
1
r

r
( +
e
)rg
g
r
,

r
r( +
e
)
g
r
,
1
r
( +
e
)g
g
r
,
1
r
2
( +
e
)g

r
(rg),
where f and g are the functions of the gasdynamic parameters f =f (, u
r
, u
z
, p, E),
and g =g(u
r
, u
z
, p). For the
NS
tensor the corresponding terms are

z
( +
e
p)
f
z
,

z
( +
e
p)g
g
z
,
1
r

r
( +
e
p)g

r
(rg),
1
r

r
( +
e
p)rg
g
r
.
The terms proportional to
e
are combined together in the vector form
FW

e
=r
_
_
_
_
_
_
J
r
D
u
r
J
r
D
+
rr
D
+
rr
NS
u
z
J
r
D
+
rz
D
J
r
D
(E + p)/ + u
r

rr
D
+ u
z

rz
D
q
r
D
+ u
r

rr
NS
_
_
_
_
_
_
, (B6)
EV

e
=
_
_
_
_
_
_
J
z
D
u
r
J
z
D
+
zr
D
u
z
J
z
D
+
zz
D
+
zz
NS
J
z
D
(E + p)/ + u
r

zr
D
+ u
z

zz
D
q
z
D
+ u
z

zz
NS
_
_
_
_
_
_
, (B7)
268 I. A. Graur and others
GG

e
=
_
_
_
_
_
0

D
+

NS
0
0
_
_
_
_
_
, (B8)
and nally the system (B1) is rewritten in the form
U
t
+
1
r
(r F)
r
+
F
1
r
+
E
z
=
1
r
(rW)
r
+
V
z

G
r
+
1
r

r
r(F
D
+ W
D
)
+

z
(E
D
+ V
D
)
G
D
r
+
1
r
(rWF

e
)
r
+
(V E

e
)
z

GG

e
r
. (B9)
The implementation of the eective relaxation time
e
for the stabilization of
numerical solution is analogous to the QGD-vector-splitting method developed and
studied by Graur (2001), and used for the solution of the QGD equations.
REFERENCES
Ashkenas, H. & Sherman, F. S. 1964 The structure and utilization of supersonic free jets in low
density wind tunnels. In Proc. 4th Intl Symp. Rareed Gas Dynamics (ed. J. H. de Leeuw),
vol. 2, pp. 84105. Academic.
Benedict, R. P. 1966 Most probable discharge coecient for ASME ow nozzles. Trans. ASME:
J. Basic Engng 734.
Bird, G. A. 1994 Molecular Gas Dynamics and the Direct Simulation of Gas Flows. Clarendon.
Chen, C. L., Chakravarthy, S. R. & Hung, C. M. 1994 Numerical investigation of separated
nozzle ows. AIAA J. 32, 18361843.
Elizarova, T. G. & Chetverushkin, B. N. 1984 Numerical algorithm for simulation of gas dynamic
ows. Dokl. Akad. Nauk. SSSR 279, 8083.
Elizarova, T. G. & Chetverushkin, B. N. 1988 Kinetically coordinated dierence schemes for
modelling ows of a viscous heat-conducting Gas. J. Comput. Math. Math. Phys. 28, 6475.
Elizarova, T. G. & Chirokov, I. A. 1999 Macroscopic model for gas with translational rotational
nonequilibrium. Comput. Maths Math. Phys. 39, 135146.
Elizarova, T. G., Graur, I. A. & Lengrand, J. C. 2001 Two-uid computational model for a
binary gas mixture. Eur. J. Mech. B/ Fluids 3, 351369.
Elizarova, T. G., Graur, I. A., Chpoun, A. & Lengrand, J. C. 1997 Comparison of continuum
and molecular approaches for the ow around a perpendicular disk. In Proc. 20th Intl Symp.
on Shock Waves (ed. B. Sturtevant), pp. 795800. World Scientic.
Elizarova, T. G., Graur, I. A., Lengrand, J. C. & Chpoun, A. 1995 Rareed gas ow simulation
based on quasi-gasdynamic equations. AAIA J. 33, 23162324.
Elizarova, T. G. & Sheretov, Yu. V. 2002 Analyse du probleme de lecoulement gazeoux dans les
microcanaux par les equations quasi hidrodynamiques. Congres de la Societe Hydrotechnique
de France (SHF), Microuidique, Toulouse, Decembre 2002, pp. 309318.
Frey, M. & Hagemann, G. 1998 Status of ow separation prediction in rocket nozzles. AIAA Paper
98-3619.
Goryainov, V. A. 2000 Origin of circulating zones in free supersonic jets. Comput. Fluid Dyn. J. 10,
401405.
Graur, I. A. 2001 Method of quasi-gasdynamic splitting for solving the Euler equations. J. Comput.
Maths Math. Phys. 41, 15061518.
Graur, I. A., Elizarova, T. G., Kudriashova, T. A., Polyakov, S. V. & Montero, S. 2002a
Implementation of underexpanded jet problems on multiprocessor computer systems. In
Parallel Computational Fluid Dynamics, Practice and Theory (ed. P. Wilders et al.), pp. 151
158. Elsevier.
Graur, I. A., Elizarova, T. G. & Lengrand, J. C. 1997 Quasigasdynamic equations with multiple
translational temperatures. Rep. R97-1. Laboratoire dAerothermique du CRNS, Meudon,
France, ISSN 0248451X.
Shock waves in expanding ows 269
Graur, I. A., Elizarova, T. G. & Lengrand, J. C. 1999 Numerical computation of shock wave
congurations in underexpanded jets. Rep. R99-2. Laboratoire dAerothermique du CRNS,
Meudon, France, ISSN 0248451X.
Graur, I. A., Elizarova, T. G., Ramos, A., Tejeda, G., Fern andez, J. M. & Montero, S. 2002b
Numerical and experimental investigation of shock waves in hypersonic axisymmetric jets. In
Shock Waves 2001, Proc. 23rd Intl Symp. on Shock Waves. Fort Worth, Texas, USA (ed. F. K.
Lu), pp. 9931001.
Graur, I. A., Ivanov, M. S., Markelov, G. N., Burtschell, Y., Valerio, E. & Zeitoun, D. 2003
Comparison of kinetic and continuum approaches for simulation of shock wave/boundary
layer interaction. Shock Wave J. 12, 343350.
Gribben, B. J., Cantarini, F., Badcock, K. J. & Richards, B. 1998 Numerical study of an
underexpanded jet. In Proc. Third Eur. Symp. on Aerothermodynamics for Space Vehicles.
ESTEC. ESA SP-426, pp. 111118.
Hirschfelder, J. O., Curtiss, C. F. & Bird, R. B. 1954 Molecular Theory of Gases and Liquids.
Wiley & Sons.
Lengrand, J. C., Chpoun, A., Graur, I. A. & Elizarova, T. G. 1995 Supersonic rareed gas ow
around a perpendicular disk. Rep. R95-6. Laboratoire dAerothermique du CNRS, Meudon,
France, ISSN 0248451X.
Liepmann, H. W. & Roshko, A. 1957 Elements of Gasdynamics. Wiley & Sons.
Mason, P. J. & Sykes, R. I. 1979 Three-dimensional numerical integration of Navier-Stokes equations
for the ow over surface-mounted obstacles. J. Fluid Mech. 91, 433450.
Mat e, B., Graur, I. A., Elizarova, T., Chirokov, I., Tejeda, G., Fern andez, J. M. & Montero,
S. 2001 Experimental and numerical investigation of an axisymmetric supersonic jet. J. Fluid
Mech. 426, 177197.
Miller, D. R. 1988 Free jet sources. In Atomic and Molecular Beam Methods (ed. G. Scoles), vol. I,
pp. 1453. Oxford University Press.
Montero, S., Mat e, B., Tejeda, G., Fern andez, J. M., & Ramos, A. 2000 Raman studies of free jet
expansion. Diagnostics and mapping. In Atomic and Molecular Beams. The State of the Art,
2000 (ed. R. Campargue), pp. 295306. Springer.
Montero, S., Ramos, A., Tejeda, G., Fern andez, J. M. & Mat e, B. 2002 Diagnostic of hypersonic
shock waves by Raman spectroscopy. In Shock Waves 2001, Proc. 23rd Intl Symp. on Shock
Waves, Fort Worth, Texas, USA (ed. F. K. Lu), pp. 425433.
Novopashin, S. A. & Perepelkin, A. L. 1989 Axial symmetry loss of a supersonic preturbulent jet.
Phys. Lett. A 135, 290293.
Parker, J. G. 1959 Rotational and vibrational relaxation in diatomic gases. Phys. Fluids 2, 449462.
Perry, A. E. & Chong, M. S. 1987 A description of eddying motions and ow patterns using
critical-point concepts. Annu. Rev. Fluid Mech. 19, 125155.
Quirk, J. J. 1994 A contribution to the greate Riemann solver debate. Intl J. Numer. Meth. Fluids
18, 555574.
Quirk, J. J. 1998 AMRITA- A computational facility for CFD modelling. Lecture Notes prepared
for VKI 29th CFD Lecture Series, 2327 February.
Ramos, A. 2001 Procesos de intercambio energ etico en expansiones supers onicas: estudio por
espectroscopa Raman. Thesis, Universidad Complutense de Madrid (in Spanish).
Ramos, A., Mat e, B., Tejeda, G., Fern andez, J. M. & Montero, S. 2000 Raman spectroscopy of
hypersonic shock waves. Phys. Rev. E 62, 49404945.
Rebrov, A. K. 1985 Free jet as an object of nonequilibrium processes investigation. In Proc. 13th
Intl Symp. Rareed Gas Dynamics (ed. O. M. Belotserkovskii), vol. 2, pp. 849864. Plenum
Press.
Rebrov, A. K. & Chekmarev, S. F. 1971 The spherical expansion of a viscous heat conducting gas
into a submerged space. Prikl. Mekh. Tekhn. Fis. 3, 122126 (in Russian).
Sheretov, Yu. V. 1997 Quasihydrodynamic equations as a model for viscous compressible heat
conductive ows. In Implementation of Functional Analysis in the Theory of Approaches. Tver
University, pp. 127155 (in Russian).
Sheretov, Yu. V. 2000 Mathematical Modeling of Gas and Liquid Flows based on Quasihydrodynamic
and Quasigasdynamic Equations. Tver University (in Russian).
Shevelev, Yu. D. 1986 Spatial Problems of Computational Aero and Hydrodynamics. Moscow, Nauka,
(in Russian).
270 I. A. Graur and others
Shevelev, Yu. D. & Klekovkin, C. G. 2003 Numerical simulation of spatial detached viscous
incompressible ows near a plate with an obstacle. J. Math. Modelling, to appear (in Russian).
Sreekanth, A. K., Prasad, A. & Prasad, D. 1991 Numerical and experimental investigations of
rareed gas ows through nozzles and composite systems. Proc. 17th Intl Symp. Rareed Gas
Dynamics (ed. A. E. Beylich), pp. 987994.
Stenholm, T. & Jover, V. 1997 Flow separation control activities at Volvo and SEP. ESA Advanced
Nozzle Workshop, University of Rome, Italy, October 1997.
Teshima, K. & Usami, M. 2001 DSMC calculation of supersonic expansion at a very large pressure
ratio. In Proc. 22nd Intl Symp. Rareed Gas Dynamics, AIP Conf. Proc. 585 (ed. T. J. Bartel
& M. A. Gallis), pp. 737744.
Volchkov, V. V., Ivanov, A. V., Kislyakov, N. I., Rebrov, A. K., Sukhnev, V. A. & Sharafutdinov,
R. G. 1973 Low-density jets beyond a sonic nozzle at large pressure drops. Zh. Prikl. Mekh.
Tekhn. Fiz. 2, 6473. English translation by Plenum Publishing Corporation, New York, 1975,
pp. 200207, UDC 532.522.2.
Welsh, F. P. 1997 Electron beam uorescence measurements of shock reection hysteresis in an
underexpanded jet. In 21st Intl Symp. on Shock Waves, Great Keppel, Australia, unpublished.

Das könnte Ihnen auch gefallen